The portfolio just got cheap. The follow-up question got expensive.

For twenty years, design hiring ran on a simple deal. The portfolio proved you did the work. The interview just confirmed the person matched the deck. That deal broke in about two years.

Now a polished case study, complete with research narrative, decision log, and before-and-after metrics, can be built in an afternoon by someone who was never near the project. The artifact stopped being proof. So the weight has to move somewhere else. Let me catch you up on where.

The proof moved, it didn't vanish

Here is the shift in one line. Vlad Derdeicea, a hiring manager writing in UX Design, puts it plainly: "The artifact is now the cheapest thing in the hiring pipeline, and the conversation is the expensive one."

The visual signal that used to mean "I can produce work at this level" now means "I have a subscription." The take-home exercise is getting dropped for the same reason. A candidate can generate it without you ever seeing them think.

What every candidate still has to do is talk to you live. So that is where the real read happens now.

The second question is the whole test

Derdeicea tells a story worth stealing. A candidate presented the strongest banking onboarding case study he had seen in months. Clean framing, real artifacts, good numbers. Then a colleague asked why they sequenced identity checks before showing any value, since teams usually do the opposite. The candidate just walked through the same screen again.

Then the killer: "What almost shipped instead?" Nothing came back. No weak answer, no answer. There was no other version, because there was no real decision behind it.

The first question is the one everyone saw coming. AI help, even live, can carry that. The follow-up that asks for the tradeoff, the regret, the specific Tuesday the call got made, that is where recall runs out. You cannot delegate having been there.

Even the live room can be gamed

Don't get smug about the interview being unfakeable. It isn't. Shraddha Sunil and Mudit Saraf, writing in Harvard Business Review after studying more than 6,000 screening sessions, found candidates can perform well with real-time AI help feeding them answers. Their line: "the ability to perform well in interviews is becoming infinitely scalable and practically free."

One caveat to hold. Both authors co-founded an interview screening company, so read that as directional evidence from people with a stake, not neutral science.

But the point stands for how you run a room. A scripted panel with predictable questions is exactly what an AI assistant handles best. Off-script follow-ups, variant problems, and "where is this weak?" are what it handles worst. Build your interview toward the second kind.

Reweight the craft you screen for

This cuts into how you check research skill too. The IxDF persona guide is a good reminder of what a defensible method actually looks like: grounded theory, real user observation over self-reporting, and triangulation to validate small samples against bigger ones. A candidate who can walk you through why they chose those steps is showing judgment. A candidate who hands you a clean persona doc is showing an output anyone can generate.

So probe the method, not the artifact. Ask how they validated a finding. Ask what they threw out. Ask where the research was thin and how they hedged the call.

The live-work format helps here. Derdeicea notes recruiters now build interviews around messy Figma files days old, decisions not yet defended. Candidates whose "articulation outruns their execution" get exposed once the conversation leaves the script.

The deep cut

Rewrite your interview kit before your next hire, not your job posts. The two questions that separate real judgment from generated polish are cheap to add and hard to fake: "What almost shipped instead?" and "Where is this work weakest?" Ask both in every loop.

And watch the tell Derdeicea flags. Candidates who name their own gaps before you find them read as credible. Ones who defend everything as perfect are performing. Train your panel to treat a good "here's what I got wrong" as a strong signal, not a red flag. That one change catches more than any take-home ever did.

Three questions for your team

1. Look at your current interview loop. How many questions could a candidate answer with a hidden AI window open, and what would you replace them with?
2. When you screen research skill, are you grading the persona doc or the method behind it? What would you ask to tell those apart?
3. Does your panel reward candidates who admit where their work is weak, or punish them? Whose scorecard needs to change?